The interest in turbulence in shallow waters, such as the Azov Sea, is caused by the fact that in the places of its existence there is an intensive transfer of the amount of motion and heat, the spread of passive impurities, the transfer of suspended particles. These processes significantly affect the formation and spatial structure of physical, chemical and biological fields of reservoirs and their spatial and temporal changes. Depending on the smoothing scale of the series of initial readings of the flow meters, the pulsation characteristics of the flow velocity were determined, which were processed in order to obtain data on turbulence and its scales. The research results provided empirical data on the conditions of generation and existence of small-scale turbulence. The collected empirical material is currently being processed to study the internal structure of the recorded disturbances of small-scale turbulence. In this paper, small-scale motion is excluded from the Navier-Stokes equations by applying the filtration operation and is modeled using subgrid models. To do this, in two-dimensional and three-dimensional cases, various types of filters are used: a box filter, a Gaussian filter and a Fourier filter, with a gradual decrease in the filter width, which allows you to reproduce a wider frequency range of fluctuations of the solution. The obtained data are planned to be used for numerical simulation of three-dimensional turbulent flows using the LES approach and comparison with the results of averaging by RANS. The article considers the possibilities of using various types of approximations for parametrization of vertical turbulent exchange. Algebraic models for calculating the coefficient of vertical turbulent exchange and semi-empirical turbulence models are compared.